The emergence of generative AI is reshaping the design of manufacturing networks. Generative AI workloads introduce large, bursty, and multiphase traffic driven by multimodal perception, while inference pipelines exhibit variable latency, high energy use, and are prone to errors. These dynamics challenge deterministic latency and reliability, required in heterogeneous factory environments where sensing, control, and robotic systems operate under strict resource constraints. This paper argues that supporting generative AI at scale requires moving beyond networks that merely use AI, toward networks that integrate AI into jointly coordinated communication, computation, and control, especially in manufacturing networks. After summarizing the challenges and opportunities of generative AI in manufacturing networks, we propose that programmable network devices can be positioned at the mobile edge, where their low-latency packet processing and lightweight in-network machine learning capabilities can directly support compute–network scheduling for 6G-enabled manufacturing. Operating at microsecond granularity, in-network machine learning offers on-path intent inference and instant scheduling signals with negligible control-plane overhead, supporting adaptive resource allocation and energy-aware workload shaping. Preliminary evaluations of heterogeneous generative AI workloads, including language and vision models, show that offloading inference to the edge is feasible and provides multiple opportunities to deploy generative AI in manufacturing networks. These results offer early evidence and practical indications for future heterogeneous scheduling strategies in next-generation manufacturing systems.
Federated Learning (FL) offers a promising approach for public, distributed, and privacy-preserving machine learning across diverse collaborative Internet of Things (IoT) environments, such as smart cities and smart healthcare. However, ensuring trust in such decentralized settings remains a critical challenge due to heterogeneous devices, limited resources, and potential adversarial attacks like model poisoning. This paper first reviews existing trust management strategies in FL, with a focus on approaches leveraging Distributed Ledger Technology, highlighting reasons, gaps, and their limitations in scalability and efficiency. Then, to overcome those issues, we propose an innovative framework that employs Directed Acyclic Graph (DAG)-based ledgers, context-aware trust indicators, and a Smart Contract to enhance trust evaluation, mitigate poisoning attacks, and ensure reliable model aggregation while maintaining scalability. Finally, we experimentally evaluate the proposed framework against state-of-the-art solutions, demonstrating clear scalability advantages as the number of participating devices increases. These results provide concrete evidence that the proposed design can serve as a practical foundation for future standardization efforts in trust management for FL.
The rapid evolution of Connected and Automated Vehicles (CAVs) is transforming the future of Intelligent Transportation Systems (ITS), where safety, reliability, and situational awareness are paramount. Achieving higher levels of automation is inseparable from Vehicle-to-Everything (V2X) communication, which extends perception beyond onboard sensors and enables 360 degrees environmental awareness. However, current V2X systems face critical limitations, including spectrum scarcity within the 5.9 GHz band, resource contention under high vehicle density, latency variability, and limited context- and relevance-aware decision-making, all of which threaten the timely exchange of safety-critical information. To address these challenges, this article explores Agentic Artificial Intelligence (AI) as a foundational enabler for semantic and context-aware communication in next-generation 6G-V2X. Operating under the Observe–Orient–Decide–Act (OODA) loop, agentic systems empower vehicles, Roadside Units (RSUs), Vulnerable Road Users (VRUs), and aerial platforms to perceive, reason, and act autonomously while coordinating collaboratively without centralized control. We propose AURA-V2X, an agentic framework for proactive and adaptive resource scheduling based on communication context and information relevance in safety-critical vehicular scenarios. Beyond presenting this framework, the paper identifies key open challenges and future research opportunities, including semantic and intent-aware communication, multi-agent coordination, sensor fusion, and secure distributed intelligence. Collectively, these advancements pave the way for resilient, context-aware, and communication-efficient transportation ecosystems, enabling scalable public safety services and supporting higher levels of vehicle automation in the 6G era.
Semantic communication, near-field operation, and integrated sensing and communication (ISAC) are often considered separately in terrestrial and non-terrestrial networks (TN–NTN). We show how a large electromagnetic aperture can connect them. Nonlocal metasurfaces and holographic apertures act on the spatial spectrum of an incident field and can execute linear projections, which form an early stage of semantic feature extraction, directly in the wave domain before digitisation, thereby reducing the corresponding digital processing burden. Nearfield focusing also provides position-dependent sensing features, allowing the aperture to serve as an analog semantic-ISAC front end on resource-constrained aerial and satellite platforms. We describe the underlying physics, the co-design of constrained analog projections with lightweight digital decoders, and a case study in which a data-adaptive projection outperforms a fully digital baseline in the simulated energy-limited regime while transmitting four times fewer coefficients and using 11.4 times fewer decoder MACs. Finally, we identify relevant 3GPP, ITU-R, and ETSI activities and the remaining standardization gaps.
Satellite-aerial-ground integrated networks (SAGINs) are envisioned to provide ubiquitous and resilient connectivity for sixth-generation (6G) networks. However, their heterogeneous and highly dynamic operating environment necessitates autonomous and self-evolving network management beyond conventional and generative artificial intelligence (AI) approaches. In this article, we investigate Agentic AI as an enabling paradigm for autonomous SAGIN through context-aware reasoning, multi-agent collaboration, and adaptive decision making. We present an Agentic AI architecture integrating perception, LLM-assisted orchestration, retrieval-augmented generation (RAG), specialized network agents, and a learning-based decision engine for closed-loop SAGIN operation. A multi-band system-level case study demonstrates its adaptability under dynamically changing network conditions. We further examine the standardization aspects of Agentic AI-enabled SAGIN in the context of 3GPP, O-RAN, and ETSI frameworks, together with practical considerations related to interoperability, computational overhead, safety, security, and trust. Finally, key research directions toward standardized and deployable autonomous SAGIN are identified.
